Update app.py

app.py

@@ -1,78 +1,86 @@
 import gradio as gr
 import os
-import json
 import faiss
-import numpy as np
+import json
 import torch
-from sentence_transformers import SentenceTransformer
+import numpy as np
 from huggingface_hub import InferenceClient, hf_hub_download
+from sentence_transformers import SentenceTransformer
 
-# 🔹 Hugging Face Credentials
-HF_REPO = "Futuresony/future_ai_12_10_2024.gguf"
-HF_TOKEN = os.getenv('HUGGINGFACEHUB_API_TOKEN')  # Ensure this is set in your environment
+# Hugging Face Credentials
+HF_REPO = "Futuresony/future_ai_12_10_2024.gguf"  # Your model repo
+FAISS_REPO = "Futuresony/future_faiss_index"  # FAISS repo
+HF_TOKEN = os.getenv('HUGGINGFACEHUB_API_TOKEN')  # Use your token
 
-# 🔹 FAISS Index Path
-FAISS_PATH = "asa_faiss.index"
+# Load Chat Model
+client = InferenceClient(
+    model=HF_REPO,
+    token=HF_TOKEN
+)
 
-# 🔹 Load Sentence Transformer for Embeddings
+# Load Sentence Transformer Model for FAISS
 embedder = SentenceTransformer("sentence-transformers/all-MiniLM-L6-v2")
 
-# 🔹 Load FAISS Index from Hugging Face
-faiss_local_path = hf_hub_download(HF_REPO, "asa_faiss.index", token=HF_TOKEN)
-faiss_index = faiss.read_index(faiss_local_path)
+# Load FAISS Index from Hugging Face
+FAISS_PATH = hf_hub_download(repo_id=FAISS_REPO, filename="asa_faiss.index", repo_type="model", token=HF_TOKEN)
+faiss_index = faiss.read_index(FAISS_PATH)
 
-# 🔹 Initialize Hugging Face Model Client
-client = InferenceClient(model=HF_REPO, token=HF_TOKEN)
+# Load FAISS Text Data
+TEXT_DATA_PATH = hf_hub_download(repo_id=FAISS_REPO, filename="asa_text_data.npy", repo_type="model", token=HF_TOKEN)
+text_data = np.load(TEXT_DATA_PATH, allow_pickle=True)
 
-# 🔹 Retrieve Relevant FAISS Data
 def retrieve_faiss_knowledge(user_query, top_k=3):
+    """Retrieve the most relevant FAISS knowledge based on user input."""
     query_embedding = embedder.encode([user_query], convert_to_tensor=True).cpu().numpy()
     distances, indices = faiss_index.search(query_embedding, top_k)
 
     retrieved_texts = []
+    print("\n🔍 DEBUG: FAISS Retrieved Indices and Distances")
+    print(indices, distances)  # 🔥 Check if FAISS is retrieving valid results
+
     for idx in indices[0]:  # Extract top_k results
         if idx != -1:  # Ensure valid index
-            retrieved_texts.append(f"Example {idx}: (Extracted FAISS Data)")
+            retrieved_texts.append(text_data[idx])  # ✅ Retrieve actual stored FAISS text!
 
     return "\n".join(retrieved_texts) if retrieved_texts else "**No relevant FAISS data found.**"
 
-# 🔹 Chatbot Response Function (Forcing FAISS Context)
-def respond(message, history, system_message, max_tokens, temperature, top_p):
-    faiss_context = retrieve_faiss_knowledge(message)
+def format_alpaca_prompt(user_input, system_prompt, history, faiss_knowledge=""):
+    """Formats input in Alpaca/LLaMA style"""
+    history_str = "\n".join([f"### Instruction:\n{h[0]}\n### Response:\n{h[1]}" for h in history])
+    faiss_context = f"\n### Retrieved Knowledge:\n{faiss_knowledge}" if faiss_knowledge else ""
 
-    # 🔥 Force the model to use FAISS
-    full_prompt = f"""### System Instruction:
-You MUST use the provided FAISS data to generate your response.
-If no FAISS data is found, return "I don't have enough information."
+    prompt = f"""{system_prompt}
+{history_str}
 
-### Retrieved FAISS Data:
+### Instruction:
+{user_input}
 {faiss_context}
 
-### User Query:
-{message}
-
 ### Response:
 """
+    return prompt
+
+def respond(message, history, system_message, max_tokens, temperature, top_p):
+    faiss_knowledge = retrieve_faiss_knowledge(message, top_k=3)  # ✅ Get FAISS data
+    formatted_prompt = format_alpaca_prompt(message, system_message, history, faiss_knowledge)
 
     response = client.text_generation(
-        full_prompt,
+        formatted_prompt,
         max_new_tokens=max_tokens,
         temperature=temperature,
         top_p=top_p,
     )
 
-    # ✅ Extract only the model-generated response
     cleaned_response = response.split("### Response:")[-1].strip()
+    
+    history.append((message, cleaned_response))  # ✅ Update history with new response
+    
+    yield cleaned_response  # ✅ Return chatbot's answer
 
-    history.append((message, cleaned_response))  # ✅ Update chat history
-
-    yield cleaned_response  # ✅ Output the response
-
-# 🔹 Gradio Chat Interface
 demo = gr.ChatInterface(
     respond,
     additional_inputs=[
-        gr.Textbox(value="You are a knowledge assistant that must use FAISS context.", label="System message"),
+        gr.Textbox(value="You are a friendly Chatbot.", label="System message"),
         gr.Slider(minimum=1, maximum=250, value=128, step=1, label="Max new tokens"),
         gr.Slider(minimum=0.1, maximum=4.0, value=0.9, step=0.1, label="Temperature"),
         gr.Slider(minimum=0.1, maximum=1.0, value=0.99, step=0.01, label="Top-p (nucleus sampling)"),